Image Stitching Technology Research

Image mosaics are the technologies for geometrically aligning and stitching two or more images of the same scene taken, at different time, form different viewpoints, and/or by different sensors, into an ultra wide-angle, seamless and high-resolution panorama. It came of photography in order to overcome the limitation of view-angle. With the development of the computer performance and image processing technologies, the automatic construction of large, high-resolution image mosaics is becoming an active area of research in the fields of photogrammetry, computer vision, image processing, and computer graphics, which has been widely used in deep space exploration, remote sensing(multispectral classification, environmental monitoring, change detection, image mosaicing, weather forecasting, creating super-resolution images, integrating information into geographic information systems (GIS)), in medicine(combining computer tomography (CT) and NMR data to obtain more complete information about the patient, monitoring tumor growth, treatment verification, comparison of the patient's data with anatomical atlases), in cartography (map updating), and in computer vision(target localization, automatic quality control), in video coding, in virtual reality, in ultra-resolution reconstruction, to name a few.The present research status and its application fields are introduced, which show us the broad allocation foreground of image mosaics. The flow of the image mosaics is presented. On the basis of it, the recent algorithms as well as the classical algorithms of the image registration techniques and the image fusion techniques, as two key techniques of the image mosaics, are introduced and discussed. The panoramic image technology, as a popular application of image mosaics, is then introduced as well as three panorama models: the cylindrical panorama, the spherical panorama and the cubic panorama. The main task of this paper is the research of block-based cylindrical panoramic image mosaics. An efficient cylindrical panoramic image mosaic technique was proposed. Firstly, the 360-degree circle panned image sequences are projected to cylindrical coordinate separately; then one image is determined as base image, and the block with most edge information is selected as the reference module, then block-matching method is used to register the neighboring images; then the transform parameters can be calculated by the registration results, finally, smooth factors are used to merge the overlapped image area data. The experimental results show that the proposed algorithm can show us the cylindrical panoramic image automatically with high robustness. Finally, a phase correlation based image sequence auto-sorting algorithm is presented in this paper. The proposed algorithm can estimate whether two images are overlapping or not, and affirms the relationship of the overlapped images. The experimental results show that the proposed algorithm is effectively and of great value.